JP2009542137A - Filter compressor and method for producing a compressed subband filter impulse response - Google Patents

Abstract

  The filter compressor (102) for generating the compressed subband filter impulse response from the input subband filter impulse response corresponding to the subband, including the filter impulse response value at the filter tap, has a higher value. In order to find a filter impulse response value from at least two input subband filter impulse responses in order to find a filter impulse response value having a value and at least one filter impulse response value having a value lower than that Processor 820 and a filter impulse response constructor (305) for constructing a compressed subband filter impulse response using the filter impulse response value having the higher value thereof, the compressed subband・ F The filter impulse response does not include a filter impulse response value corresponding to a filter tap of at least one filter impulse response value having the lower value, or at least one filter impulse having the lower value. Contains a zero value corresponding to the filter tap of the response value.

Description

  The present invention provides a QMF domain (QMF = orthogonal mirror filter bank) that can be used, for example, in the field of speech applications such as filtering of the head related transfer function (HRTF) for multi-channel sound experiences on headphones. ) In the subband domain, sometimes referred to as).

  Recent developments in filter transformation technology have enabled a very efficient QMF representation of time domain filters. In general, any FIR filter in the time domain (FIR = finite impulse response) can be transformed into a set of complex filters, each corresponding to a particular subband in QMF. Thus, filtering can occur in the complex QMF domain, similar to the way that filtering can be performed using FFT (FFT = Fast Fourier Transform). Nonetheless, the QMF domain representation of filtering and the computational complexity of implementation can be substantial, for example for filters with long impulse responses in the time domain.

  Furthermore, recent developments in speech coding have made available the ability to reproduce multi-channel representations of speech signals based on stereo (or monaural) signals and corresponding control data. These methods transmit additional control data to control reproduction, also called upmix, of surround channels based on the transmitted mono or stereo channel, so Dolby Prologic® It is very different from the old matrix based solutions like)).

  Thus, such a parametric multi-channel audio decoder, eg MPEG Surround (MPEG Surround®), reconfigures N channels based on M transmitted channels and further control data, N and M are possible integers, N> M. This additional control data represents a significantly lower data rate than all transmissions of the N channels, making encoding very efficient while ensuring compatibility of both M-channel and N-channel devices.

  These parametric surround coding methods typically involve the parameterization of surround signals based on IID (Interchannel Intensity Difference) and ICC (Interchannel Coherence). These parameters represent the power ratio and correlation between channel pairs in the upmix process. Additional parameters that are also used in the prior art include prediction parameters used to predict intermediate or output channels during the upmix procedure.

  Other advances in speech coding have provided a means to obtain a multi-channel signal impression on stereo headphones. This is generally done by downmixing the multichannel signal to stereo using the original multichannel signal and a so-called HRTF (head related transfer function) filter. A parametric multi-channel audio decoder that first reproduces the multi-channel signal from the transmitted down-mix signal and then renders the multi-channel signal on the headphones without having to down-mix it again with an HRTF filter. It has been shown in the prior art that it can be combined with the binaural downmix algorithm that enables it. This is accomplished by combining the HRTF filter into four filters as a function of the parametric multi-channel representation. As a result, how the four filters are combined to achieve a binaural or stereo output signal (2 channels) resulting in a stereo signal (2 channels) used as input for multi-channel representation or Expressed as a function of parametric multi-channel representation of what is mixed. Thus, each of the four filters is associated with one of the two input signals with respect to the two output signals. However, HRTF filters can be very long to model room characteristics well, so the computational complexity for filtering four HRTF filters in the QMF domain can be significant.

  According to an embodiment of the present invention, a filter for generating a compressed subband filter impulse response from an input subband filter impulse response corresponding to a subband, including a filter impulse response value at a filter tap. The compressor filters from at least two input subband filter impulse responses to find a filter impulse response value having a higher value and at least one filter impulse response value having a value lower than the higher value. A processor for examining the impulse response value and a filter impulse response constructor for constructing a compressed subband filter impulse response using the filter impulse response value having the higher value, the compressed subband ·fill The impulse response does not include a filter impulse response value corresponding to a filter tap of the at least one filter impulse response value having the lower value, or at least one filter impulse response having the lower value Contains a zero value corresponding to the value filter tap.

  A further embodiment of the present invention relates to a method for creating a compressed subband filter impulse response from an input subband filter impulse response corresponding to a subband, including a filter impulse response value at a filter tap, In order to find a filter impulse response value having a higher value and at least one filter impulse response value having a value lower than the higher value, the filter impulse response value from at least two input subband filter impulse responses And using the filter impulse response value having the higher value to construct a compressed subband filter impulse response, the compressed subband filter impulse response having the lower value Having at least one A zero value corresponding to the filter tap of at least one filter impulse response value that does not include or has a lower value than the filter impulse response value corresponding to the filter tap of the filter impulse response value of Including.

  Embodiments of a computer readable storage medium include multiple sets of subband filter impulse responses stored thereon, each set of subband filter impulse responses together being a time domain head related transfer function related filter And the filter impulse response of the time domain head related transfer function related filter is greater than the sum of the lengths of the subband filter impulse responses of each set of subband filter impulse responses, or The filter impulse response of the time domain head related transfer function-related filter is the complex number of the subband filter impulse response of each set of subband filter impulse responses when the filter impulse response values are complex values. Value filter imper Greater than the sum of the length of the scan response value.

  Some embodiments of the invention may be advantageous with respect to balancing computational efficiency on the one hand and quality on the other hand. Embodiments provide both a significant reduction in computational complexity and a good approximation of the filter represented by the input subband filter impulse response. Examining (and eventually including selecting or determining) and configuring the compressed subband filter impulse response with the selected (or determined) filter impulse response value may include several Both reduction in computational complexity and good approximation can be achieved in embodiments and / or applications, leading to (almost) audio indistinguishable listening experiences. In some embodiments, this can be accomplished by finding, selecting or determining a filter impulse response value of an input filter impulse response having a higher value, while at least one filter The impulse response value is not selected or determined and has a value lower than the higher value. Using the selected or determined filter impulse response value or a filter impulse response value having a higher value, a compressed filter impulse response having a compressed filter impulse response value is constructed or created. Depending on the implementation, filter impulse response values that are not selected or not determined or that have a value lower than a higher value are set to zero or ignored. In other words, the filter impulse response value may include an ignored pattern, set to zero, or otherwise modified filter impulse response value.

  Further, some embodiments provide a wide range of achievable reductions in computational complexity by influencing the selection of filter impulse response values from which compressed subband filter impulse responses are constructed. Can do. As a result, some embodiments of the present invention provide tremendous flexibility in balancing the achievable fit of computational complexity on the one hand and the approximate quality on the other hand.

  Thus, some embodiments of the invention can be applied particularly in the field of speech or other applications that include filters with relatively long (finite) impulse responses in the time domain. By transforming the filter or filter element from the time domain to the (complex) subband domain, as will be explained later, the calculation is such that the impulse response of the individual subband filter is the filter impulse response of the filter in the time domain. It can be performed in parallel so as to be significantly shorter than the comparison.

  However, the overall computational complexity cannot be reduced by a simple transformation from the time domain to the (complex) subband domain. For example, for a filter with a relatively long impulse response, such as an HRTF filter, even an individual subband filter usually has a long finite impulse response, which is divided by the number of individual subbands. The order of the finite impulse responses of the corresponding filters is very roughly described. Thus, depending on the computing capabilities available in a particular application, the overall computational complexity or even the computational complexity for individual subband filters can be substantial.

  Furthermore, additionally or alternatively, a determination based on the level of the filter impulse response can be implemented in a filter compressor embodiment. In such a case, the filter compressor may have at least one filter impulse response value set to zero or ignored if the value of the filter impulse response (eg absolute value) is below a threshold. Can be configured. In some areas of the application, the one or more filter impulse response values may be close to the aliasing level of the filter bank corresponding to the input subband filter impulse response. If the value of the filter impulse response value is close to the aliasing level of such a corresponding filter bank, a particular tap can be set without any problem to the corresponding filter coefficient or filter impulse response value to zero. May be set to zero. As a result, implementations of filters based on such compressed filter impulse responses do not have to perform multiplication and addition for zero valued coefficients or impulse response values.

  In this connection, the filter bank aliasing level is an inherent characteristic of many filter banks. Such an aliasing level of the filter bank arises from mere processing of the signal, for example in the framework of SBR applications. Since each filter tap or filter impulse response value contributes to the output producing the signal, the tab (eg absolute value thereof) is smaller and the result or contribution of each tap is smaller with respect to the output of the filter bank. Thus, small taps will have such a small contribution to the output of the order of the aliasing levels of the filter banks or the respective filter banks whose contributions are in that range. In this case, further distortion introduced by setting the corresponding tap to zero can often be tolerated so as not to introduce further audible distortion. In many cases, typical ranges of aliasing levels are in the range of −30 dB, −40 dB, −50 dB, −60 dB, and −70 dB and lower compared to the peak signal.

  For example, in the case of an HRTF filter, after converting the time domain HRTF filter to a complex QMF representation, some of the time frequency tiles in the complex QMF representation have a low absolute value (at the aliasing level of the MPEG Surround filter bank). obtain. These entries in the complex QMF representation of the HRTF filter can be set to zero. This allows for reduced complexity for implementing a long HRTF filter with room response included in the complex QMF representation. Thus, to achieve binauralization with reduced complexity, realistic filter effects can be maintained and the filter converter can be followed by a filter reduction process in the form of a filter compressor embodiment. The filter reduction step aims to simplify the HRTF filter so that the subband HRTF filter contains at least a small number or even a significant number of zeros. Since fewer coefficients are active, a significant reduction in computational complexity can thereby be achieved.

  Thus, a computer comprising a set of compressed subband filter impulse responses and a plurality of compressed subband filter impulse responses provided by an embodiment of a filter compressor, an embodiment of a method for making it Embodiments of readable storage media can significantly reduce the individual computational complexity for each subband filter and the overall computational complexity for all subband filters.

  The present invention will now be described by way of example with reference to the accompanying drawings, which do not limit the scope or spirit of the invention.

FIG. 1 illustrates the interaction of an embodiment of a filter converter and a compressor according to the present invention. FIG. 2 illustrates an example scenario for the present invention. FIG. 3 illustrates an embodiment of a filter compressor according to the present invention. FIG. 4 illustrates a further embodiment of a filter compressor according to the present invention. FIG. 5 illustrates a further embodiment of a filter compressor according to the present invention operating simultaneously on multiple filters. FIG. 6 illustrates an embodiment of the present invention used in the context of HRTF filtering. FIG. 7 illustrates a possible solution for a compatible filter. FIG. 8 illustrates a possible solution for the main components of the filter converter. FIG. 9 illustrates a possible solution for a (complex) analysis filter bank. FIG. 10 illustrates a possible solution for a compatible subband filter bank. FIG. 11 illustrates a first possible solution of the (complex) synthesis filter bank. FIG. 12 illustrates a second possible solution for a (complex) synthesis filter bank. FIG. 13 illustrates a further embodiment of a filter compressor according to the present invention. FIGS. 14a-14c illustrate spectral whitening as used in a filter compressor embodiment according to the present invention. FIG. 15 illustrates an embodiment of a filter compressor according to the present invention that operates simultaneously on multiple filters.

  The embodiments described below are merely illustrative for the principles of the present invention for efficient filter representation. It will be understood that changes and modifications in the arrangements and details described herein will be apparent to those skilled in the art. Accordingly, it is intended that the invention be limited not by the specific details presented as the description and description of the embodiments herein, but only by the forthcoming claims.

  Before describing embodiments of the present invention, further components and applications of the invention in more detail, it should be noted that objects, configurations and components having the same or similar functional characteristics are indicated by the same reference signs. Unless expressly noted otherwise, descriptions of objects, configurations and components having similar or equivalent functional characteristics and features may be interchanged with respect to one another. In addition, in the following, unless specific characteristics, configurations and characteristics or features of components are stated, the objectives that are the same or similar in one embodiment or appear in different configurations shown in one of the figures, Summarizing configuration and component citations is used. By summarizing the reference signs, a more compact and clearer description of the embodiments of the invention is possible, and the possibility of exchanging features and descriptions between the different embodiments is clearly shown.

  Furthermore, it should be noted that in the following, the embodiments shown in the figures equally represent corresponding embodiments of the method. Thus, the embodiment shown in the figure not only illustrates a corresponding embodiment of a filter compressor, for example, but also represents a flowchart of a corresponding embodiment of the corresponding method. As outlined below, such embodiments of the method can be implemented in hardware or in software.

  In FIG. 1, one embodiment of the present invention is outlined along with possible applications. Specifically, FIG. 1 shows a filter converter 101 connected to an embodiment of the filter compressor 102. The filter converter 101 will be described in more detail later. An embodiment of the filter converter 101 is provided with an input signal that contains information about the finite impulse response h (n) of the filter or filter element in the time domain. The index n is an integer indicating a different value or sample of the finite impulse response (FIR) in this connection, and h (n) is a real-valued number.

  The finite impulse response of the time domain filter h (n) is the filter or filter element response in the time domain on the excitation in the form of a single impulse having a predetermined amplitude. In principle, the complete behavior of the filter element in the time domain is included in the finite impulse response of the filter. For a digital system, the impulse response of the filter can be determined or measured by applying an input signal having a value different from zero in a single instance in time. This value may be equal to 1, for example.

  The filter converter 101 can provide a set of finite impulse responses H (n, k) that can be used in a compatible filter framework, as outlined in connection with FIG. Note that for complex filter converters based on complex analysis filter banks, the finite impulse response H (n, k) contains a number of complex values, where n represents a different sample as before, k = 0,..., (L−1) indicates the corresponding subband to which the finite impulse response of the subband filter corresponds. Both l and k are integers. Furthermore, the number L of subbands is also a positive integer. For digital systems, the number L of subbands provided by the filter converter 101 and used later for the filtered digital audio input signal is often a power of 2, eg 16, 32, 64, 128, 256, 512. In the following example, the number of subbands is selected such that L = 64. However, as outlined above, in principle, any positive integer L can be used as the number of subbands in filter compressor applications, components and embodiments.

  As described, the time domain filter h (n) is input to a filter converter 101 that produces a complex QMF or subband representation of the filter H (n, k). In this particular example where L = 64 subband QMFs are used, the complex QMF representation of the filter is the length for a length K time domain filter with a number of L = 64 finite impulse response lengths. It is represented by an L = 64 complex filter of K / 64 + 2.

  In accordance with the present invention, the filter H (n, k) is then input to the filter compressor 102, which outputs H (n, k) as a compressed subband filter impulse response. Embodiments of the filter compressor 102 output a filter H (n, k) having a greater number of zero-valued coefficients than the original filter H (n, k), thus allowing for low computational complexity.

  Depending on the embodiment and application, the filter converter 101 and the filter compressor 102 are coupled to each other via an L connection, each with a different subband (index k = 0,..., L−1 or k = 1). ,..., L), the filter impulse response is transmitted. This option is indicated in FIG. 1 by a slash (/) crossing the connection of the filter converter 101 and the filter compressor 102. However, the two components may be coupled to each other by a fewer number of connections or only by a single connection, and corresponding signals or information are transmitted. For simplicity of the figures and embodiments shown, possible parallel connections of elements including individual connections for each subband are shown as needed. However, whenever a signal or information about a subband is transmitted, any connection can be made, for example as indicated by a variable indicating it (eg H (n, k)).

  As will be described in more detail later, embodiments of the filter compressor 102 also output one or more filter impulse responses for each number of subband filters included in the subband filter bank, for example. To do. Both the input subband filter impulse response H (n, k) and the compressed subband filter impulse response H (n, k) are in time-related n and subbands as previously described. The number of both complex values arranged in a two-dimensional matrix represented by the associated k.

  However, further details regarding different embodiments of the filter compressor 102 will be outlined later. Further, the relationship between the compressed subband filter impulse response H (n, k) and the input subband filter response H (n, k) will be described later for different embodiments of the filter compressor 102. In principle, the two sets of multiple filter impulse responses H (n, k) and H (n, k) can differ in more ways by the number of zero-valued coefficients, as briefly outlined. It is important to keep in mind.

  In FIG. 2, a general use case scenario for the present invention is outlined. Here, the time domain filter h (n) is input to a filter converter 101 that produces a complex QMF representation of the filter H (n, k), which is described previously. As such, it is input to an embodiment of the filter compressor 102 that outputs a reduced or compressed complex QMF filter H (n, k).

  Apart from the embodiment of the filter converter 101 and the filter compressor 102, which both provide the real-valued impulse response of the filter in the time domain h (n) described in connection with FIG. 1, the use shown in FIG. The example scenario further includes a QMF analysis filter bank 203, also referred to as a complex analysis filter bank. The QMF analysis filter bank 203 is provided with an input signal x (n), which can be, for example, a digital audio signal. The QMF analysis filter domain 203 provides a complex QMF representation X (n, k) of the input signal x (n) at the output. As described in connection with FIG. 1, integers n and k are associated with a sample or time index and a subband index, respectively. A possible solution for the QMF analysis filter bank 203 is described in more detail in connection with FIG.

  The complex QMF representation X (n, k) of the input signal x (n) is then provided to a filtering stage 201 that operates in the subband domain. Filtering stage or subband filter 201 is an adjustable subband filter bank, which includes a plurality of L intermediate filters coupled to the output of an embodiment of filter compressor 102. Through an embodiment of the filter compressor 102, the intermediate filter of the subband filter bank 201 includes a compressed subband filter used to filter the (complex-valued) QMF representation X (n, k). An impulse response H (n, k) is provided.

  In principle, as will be explained later, the complex QMF representation X (n, k) is a complex QMF representation X (n, k) and filter compressor for each subband identified by the subband index k. It can be filtered by calculating the convolution of each filter impulse response H (n, k) provided by the 102 embodiment.

  Then, the filter signal provided by the subband filter bank 201 in the complex QMF domain is provided to the QMF synthesis filter bank or the complex synthesis filter bank, which finally outputs the (real-valued) output signal y (n). Synthesize. Possible solutions for the QMF synthesis filter bank 202 or the complex synthesis filter bank are described in the framework of FIGS.

  In other words, as shown in FIG. 1, in parallel with the embodiment of the filter converter 101 and the filter compressor 102, the signal x (n) is represented by X (n, k), ie the combined complex QMF representation of the input signal. It is input to the QMF analysis 203 module to be output. The signal is then applied to filter 201 in the QMF domain using a complex QMF filter output by filter compressor 102, and the filtered signal produces a filtered output signal y (n). It is finally synthesized into the time domain by the bank 202.

  In FIG. 3, a more detailed view of an embodiment of the filter compressor 102 is provided. Also, the time domain filter h (n) as an input impulse response in the time domain is input to the filter converter 101. The time domain impulse response of the filter is indicated by 301. As explained previously, after the filter converter, the time domain filter is forwarded to the subband domain and is represented by H (n, k). A time / frequency plot of the absolute value of the filter response is given by 302.

  The embodiment of the filter compressor 102 shown in FIG. 3 includes an absolute value representation module 303 that is connected to the input of the embodiment of the filter compressor 102. The filter compressor 102 embodiment further includes a mask generator 304, which is coupled to the output of the absolute value representation module 303. The filter calculator 305 is also included in the filter compressor 102 embodiment, which is connected to both the input of the filter compressor 102 embodiment and the output of the mask generator 304. The filter calculator 305 also includes an output, which represents the output of the filter compressor 102 embodiment.

  The complex QMF filter or subband input filter H (n, k) is input to an embodiment of the filter compressor 102 that includes an absolute value representation module 303, a filter mask generator 304 and a filter calculator 305. The absolute value representation module 303 creates a time / frequency plot of the absolute value of the filter, as illustrated by the partial diagram 302. This can be, for example, a logarithmic representation of the absolute value of the filter coefficient in the QMF domain, as outlined later. The filter mask generator 304, in one embodiment, selects the coefficient (n, k) having the maximum value in the representation of the absolute value of the filter in the QMF domain based on the information provided by the absolute value representation module 303. Or decide. The filter mask generator 304 determines or selects an adjustable, programmable, constant or predetermined number of coefficients depending on the amount of filter compression required. A lower number of selected filter coefficients gives a higher complexity reduction. Examples and further details are further described in the specification. Thus, in many cases in the framework described herein, determining, selecting, establishing, establishing and finding words can be used interchangeably. In many cases, the determined or selected filter impulse response value is such a filter impulse response value, which is high compared to a filter impulse response value having a lower value than a higher value. Has (or contains) a value. These low value filter impulse response values are considered as not being selected or determined.

  As outlined previously, alternatively or additionally, the complexity reduction is compared to the so-called aliasing level of the filter bank corresponding to the filter impulse response, as provided in the filter compressor embodiment. It can also be achieved based on examining the filter tap or filter impulse response value. If certain taps of the filter impulse response value in the QMF domain are close to the aliasing level of the filter bank, these filter taps are set to zero or otherwise handled without difficulty to reduce computational complexity Can be. These filter taps can be safely ignored in the filter implementation so that zero-valued coefficients do not need to be included in the multiply-add framework in the filter implementation. For example, after converting a time domain HRTF filter to a complex QMF representation, some of the time frequency tiles in the complex QMF representation may have a low absolute value at the aliasing level of the corresponding MPEG Surround filter bank. These entries in the complex QMF representation of the HRTF filter can be set to zero, which allows for reduced complexity for implementing long HRTF filters with an included room response.

  The filter mask generator creates a filter mask M (n, k) based on the information provided by the absolute value representation module 303, and further selects the selected filter coefficient of H (n, k) as a filter calculator. The selected filter mask M (n, k) shown at 305 is output. The filter calculator 305 creates a new compressed filter H (n, k) from the original filter H (n, k) in the QMF domain that includes the selected filter coefficients. Further details on the different possibilities for implementation are given below.

  FIG. 4 shows a further embodiment of the filter compressor 102, which has the same basic configuration as the embodiment of the filter compressor 102 shown in FIG. Specifically, the embodiment of the filter compressor 102 shown in FIG. 4 also includes an absolute value representation module 303, which is connected on the one hand to the input of the embodiment of the filter compressor 102 and on the other hand the absolute value representation module. It is connected to the mask generator 304 via the output of 303. The embodiment of the filter compressor 102 in FIG. 4 also includes a filter calculator 305 that is connected to the input of the filter compressor and the output of the mask generator 304. The output of the filter calculator 305 represents the output of the embodiment of the filter compressor 102 shown in FIG. 4 as before.

  However, compared to the embodiment of the filter compressor 102 shown in FIG. 3, the absolute value representation module 303 and the filter calculator 305 are shown in more detail in the case of the embodiment shown in FIG. Further details are given in the following sections along with alternative or further implementations.

  The absolute value representation module 303 includes an absolute value algorithm function module 401, which is connected in series with the whitening module 402 between the input and output of the absolute value representation module 303. The filter calculator module 305 includes a filter decimator module 403 that is connected in series with a gain calculator 404. Both the filter decimator module 403 and the gain calculator 404 are connected in series between the input and output of the filter calculator module 305. Depending on the specific implementation, information about the mask as provided by the mask generator 304 is provided to the filter decimator module 403 and optionally to the gain calculator module 404 as shown in FIG. The However, depending on the specific implementation of the filter calculator module 305, the gain calculator module 404 may be passed through any connection between the inputs of the gain calculator module 404 and the filter calculator module 305, as shown by the dashed lines in FIG. Thus, as provided in the embodiment of the filter compressor 102, an input subband filter impulse response H (n, k) may optionally be provided.

  Before discussing the individual modules of the filter compressor 102 embodiment shown in more detail in FIG. 4, a summary of the functionality of the embodiment of the filter compressor 102 is given as shown in FIG.

  In FIG. 4, a different embodiment of the filter compressor 102 according to the present invention is outlined. Here, the absolute value representation module 303 includes an absolute value and logarithmic function module 401 and a whitening module 402 that performs spectral whitening of the absolute value representation supplied by the absolute value and logarithmic function module 401. The filter mask generator 304 is the same as before, and outputs the filter mask M (n, k) to the filter calculator module 305. This is because the filter decimator module 403 holds the selected coefficients of the filter H (n, k) and sets the other coefficients to zero in this embodiment, and the gain of the compression filter H (n, k) And a gain calculator module 404 that adjusts the gain of the filter to be similar to that of the original filter H (n, k).

  The evaluation expression A (n, k) shown by equation (1) reflects the volume allocation for the human ear without taking into account the specific acoustic properties of the human ear.

  The partial diagram 302 in FIG. 3 schematically representing the evaluation expression A (n, k) is a three-dimensional plot of the evaluation expression A (n, k) as a function of the two indices k, n in the plane shown in FIG. Further, the evaluation expression value A (n, k) is plotted at right angles to the nk plane of the partial view 302. In other words, the partial diagram 302 shows a schematic representation of the evaluation representation of the time / frequency representation of the absolute value of the filter A (n, k) as a function of the sample index or time index n and the subband index k. The time index or sample index n may differ, for example, by an index n to L (number of subbands) times the time domain impulse response h (n). As described in connection with FIGS. 9, 11 and 12, the filter converter 101 may include a complex modulated analysis filter bank, which in turn may include one or more downsamplers, It reduces the number of samples by a factor of times, which can be, for example, the number of subbands L. However, as these downsamplers are optional components, the index n refers to a time index or sample index corresponding to the index of the time domain impulse response H (n), or, for example, It may correspond to a down-sampled time index or sample index that differs by L times from the time or sample index n of the impulse response H (n).

  In the following, further details about the whitening module 402 are outlined. The purpose of the whitening module taught by the present invention is to allow perceptual weighting of filters prior to mask generation to avoid situations where perceptually important filter taps are discarded. The reason is that they have small absolute values due to other perceptually less important filter taps.

  Even though spectral whitening is described in more detail in particular with regard to whitening as represented by equation (4) in connection with FIG. 14, (spectral) whitening is It should be noted that it is based on the finding that it is desirable to transfer energy from one spectral part to another in order to prevent or minimize the distortion created in the process.

  Real-world filters and audio systems very often have an irregularly distributed time / frequency distribution that is located at lower frequencies rather than subbands located at higher frequencies. A filter impulse response can be provided in the subband domain that has a significantly greater length compared to the subband being. Furthermore, the random distributed amplitude / frequency distribution of real filters and audio systems can also lead to different associations of individual subband filters with respect to each other. In other words, a subband filter corresponding to a higher frequency is compared to a subband filter corresponding to a lower frequency, for example due to higher attenuation than real filters and audio systems at higher frequencies. And not so important. However, to prevent or at least minimize the effect that filter compression has on higher frequency subband filters, (spectral) whitening is completely suppressed in the process of compression, leading to intense distortion of the listening experience. Thus, subband filters at higher frequencies can be advantageously implemented to prevent in the scenario outlined above. Thus, (spectral) whitening, also called weighting, can be an important point for real-world filters and audio systems.

  Therefore, as included in the absolute value representation module 303 in the embodiment shown in FIG. 4, the whitening module 402 applies spectral whitening, where the normalization effect is used by dividing the entire frequency range into frequency bands. To do. Each subband corresponds to a specific frequency range having a specific center frequency, as will be described in more detail in the context of a complex modulated analysis filter bank. As a result, the subbands can be arranged according to the center frequency. In a natural choice, the subband index k corresponds to an increasing order to the center frequency in increasing order.

  To perform spectral whitening in the form of a normalization effect with respect to the aforementioned frequency bands, perceptually related intervals of subbands or subgroups of subbands are formed, each including at least one subband. Further, in many specific implementations, individual subbands belong to exactly one subgroup as a whole. However, each subgroup of subbands may include multiple subbands. In this case, the subgroup typically includes only subbands with adjacent center frequencies.

  In other words, if the subbands are arranged according to a subband index k that increases further simultaneously according to their center frequencies in increasing order, the subgroup containing only subbands with adjacent frequencies is subband indexed. associated with subbands having k, which is arranged such that the maximum difference between two arranged subband indices is equal to +/− l as described in connection with equation (3) Can do. In other words, each frequency band can be represented by a subgroup or interval of subbands, which is a superset of subbands. However, it should be noted that a subgroup of subbands may contain exactly one subband.

  As mentioned above, in the spectral whitening framework, a certain number P of frequency bands, subband subgroups or intervals are distinguished. In principle, the number p of subgroups of subbands is an integer, which means that each subgroup contains at least one subband, and each subband belongs to exactly one subgroup of subbands. Therefore, it is smaller than the number L of subbands. For a filter system operating in L = 64 subbands, a typical number P of subband subgroups can be selected to be 28. However, this number is not limited as described above. A corresponding number P of subgroups of subbands (eg, P = 32) can be selected based on a psychoacoustic model representing perceptually relevant intervals in the frequency domain.

  Thus, whitening leads to a transfer from a lower spectral portion of energy to a higher spectral portion, in many real filters and audio systems, optionally based on the perceptual characteristics of the human ear with respect to psychoacoustic models.

  However, different implementations of the whitening module 402 can be easily implemented in the framework of the absolute value representation 303. Specifically, other implementations may be performed for each subband with index k instead of performing whitening based on all subbands included in each subgroup of subbands according to equation (4). The evaluation expression A (n, k) is individually whitened. Further, instead of subtracting the maximum value as shown in equation (4), whitening divides all values of the evaluation expression A (n, k) and thereby with respect to the maximum of each subband or sub This can be done by normalizing all values of the evaluation expression with respect to the maximum value of each subgroup of the band. Furthermore, the normalization described by dividing the evaluation expression is such that the sum of all values of each evaluation expression A (n, k) is (for each individual subband or for each subgroup of subbands). It can also be performed just as it can. In this case, in a first step, the sum of all values of the evaluation expression is determined for each subband or for each subgroup of subbands, followed by subtraction according to equation (4) or the respective sum value. Followed by splitting the value of the evaluation expression having.

  In summary, in the embodiment outlined above, examining and selecting is based on the absolute value of the filter impulse response value at the filter tap. Thus, in this embodiment, the filter impulse response value is selected or not selected based on a comparison on the absolute value of the filter tap when selecting at least one that includes a higher value. In different embodiments, comparing or examining the filter taps as needed may be based on the application of other mathematical means. If the filter tap is real-valued, in principle, no mathematical means need be applied, but calculating or determining the absolute value may be implemented.

  In the case of complex-valued filter taps, it is desirable to apply some mathematical means. Examples may be deriving absolute values or deriving the filter tap angle or phase with respect to a predetermined or unambiguous direction (eg positive real direction) in a complex plane. In addition, determine the real part, the absolute value of the real part, the imaginary part, the absolute value of the imaginary part, or other functions that map each complex number to a (optionally positive) real number as a rule. Can do.

In the embodiment shown in FIG. 4, the whitened evaluation expression A _W (n, k) as output by the whitening module 402 is provided to the mask generator 304, which is based on the whitened evaluation expression. Create a filter mask or mask M (k). Due to the actual whitening module 402 for the evaluation representation, the mask generator 304 can select the most (perceptually) relevant filter coefficients. The filter mask is a set of 0s and 1s in the embodiment shown in FIG. 4, where M (n, k) = 1 is used or retained with the corresponding filter tap or filter impulse response value. To be selected. Thus, the value M (n, k) = 0 is not used because the corresponding filter tap or filter impulse response value identified by the sample index or time index n and the subband index k is not selected. Indicates that it is not possible. In other words, the particular filter impulse response value is ignored or set to zero.

The specific implementation of the mask generator 304 may vary greatly depending on the embodiment of the filter compressor 102. In the embodiment shown in FIG. 4, the mask generator sets, for example, the corresponding value of the filter mask M (n, k) = 1, but the remaining value of the filter mask is set to 0 A particular number of impulse response values can be selected based on the whitened evaluation expression A _W (n, k). Apart from selecting a specific absolute number of impulse response values, relative numbers for the entire number of impulse response values given by a set of subband filter responses H (n, k) are also possible. In a specific example in the case of an L = 64 QMF subband implementation where each input subband filter impulse response includes 16 non-zero, non-zero or non-trivial filter taps, The entire matrix of input subband filter responses is given by a 64 · 16 matrix containing 1024 impulse response values. In this example, the mask generator 304 may select, for example, a particular predetermined number of impulse response values (eg, 256 elements with the largest absolute value as provided by the whitened evaluation representation). The mask generator 304 can select a predetermined or specific ratio (relative number) of filter impulse responses (eg, 25% of the total number of filter response values) with respect to the total number of filter impulse responses. be able to. In any case, the remaining impulse response values are ignored or selected by setting the corresponding value of the filter mask M (n, k) equal to zero (M (n, k) = 0). Not.

  In a further embodiment of the filter compressor 102, the mask generator 304 is configured to receive a signal indicating the selected number of impulse response values or indicating the ratio of the impulse response values with respect to the total number of impulse response values. May be. In such an embodiment of the filter compressor 102, the compression ratio can be adjusted by adjusting the previously mentioned figures.

  Further, the mask generator 304 may alternatively or additionally be configured to select each filter impulse response value based on different criteria. For example, the mask generator 304 may determine a predetermined, constant, programmable or compatible number of impulse response values for each subband (eg, three impulses having a maximum value for the evaluation representation for each subband. Response value) may be selected. Further, the mask generator 304 holds a corresponding evaluation expression value that is greater than a predetermined, constant, adjustable or programmable threshold, for example, so that all impulse response values are selected. May be configured. In a further embodiment, it may be desirable to configure the mask generator 304 such that impulse response values can be selected based on a comparison with the respective values of adjacent impulse response values. For example, the mask generator 304 may have a value that is less than a predetermined, programmable, or adjustable ratio that is constant relative to adjacent values taking into account (optionally whitened) evaluation expressions ( If, for example, less than 25%, the filter impulse response value may not be selected. However, other selection schemes can be implemented.

  However, for whitening, as described in connection with equation (4), based on each subgroup of subbands or individual subbands, at least one impulse response value is selected from the selected impulse response. The number of values varies greatly from subband to subband or from subgroup to subband, but is selected in each subgroup of subbands or in each subband, depending on the specific implementation. For example, in the case of whitening implemented by dividing the evaluation expression A (n, k) by the maximum value of the corresponding subset of evaluation expression values, in the above implementation of the mask generator 304, at least one filter impulse. The response value is selected in each subband or in each subgroup of subbands, as described in connection with FIG.

  As a result, the interaction between the absolute value representation module 303 and the mask generator 304 is the set of important values of the filter impulse response value in the nk plane (see the partial diagram 302 in the reference FIG. 3) and the n− It leads to “decompression” or “compression” of “air” between perceptually related areas of the k-plane. Therefore, the associated impulse response value is ignored by setting the mask M (n, k).

  In this embodiment, the mask consists of entries that are zero or one. An entry with a zero indicates which filter coefficients are to be discarded, and an entry with a 1 indicates which filter coefficients are to be retained (selected).

In any case, G _max is an upper bound for gain compensation, and ε is a small positive number included to avoid division by zero. Thus, both G _max and ε are numbers, and they prevent division by zero, such as by the minimum of two terms in parentheses bracketed in equations (6) and (8) ( I.e., ε> 0), and further, to limit the gain applied to the subbands by the gain calculator module 404 to a value as defined by the maximum gain G _max , the numerical calculation of the gain calculator 404 is useful for each Gains G (k) and G (p) are limited to the value of G _max .

Accordingly, gain calculator module 404 normalizes the filter taps H _M (n, k) masked with respect to energy to compensate for energy lost in the process of masking at least some input subband input responses. In other words, for masking in the filter decimator 403 framework, the signal filtered by the subband filter input response corresponding to the masked subband filter impulse response H _M (n, k) is subband Has a smaller energy compared to a subband filter using a filter impulse response H (n, k).

  However, the gain calculator module 404 may be configured to apply different gain schemes. As an example, a direct comparison of the absolute values of the subband filter impulse response rather than energy can be used to determine the gain factor. Additionally or alternatively, the gain factor G may be an overall rather than an impulse response value for individual subbands or individual subgroups of subbands, as described in connection with equations (6) and (8). Can be determined based on the number of subband filter impulse response values. Furthermore, it should be noted that the gain calculation module 404 is not a required component but rather an optional component.

  The filter impulse response constructor or filter calculator module 305, in a further embodiment of the present invention, not only by setting the unselected subband filter impulse response value to zero as described above, A filter impulse response can be constructed. Depending on the specific implementation, the filter impulse response constructor 305 weights the appropriate selected or determined subband filter impulse response values, eg, to construct a compressed subband filter impulse response. This can be accomplished by copying, or obtaining.

  In this connection, it should be noted that even if ignoring or not including a filter impulse response value that has not been determined or selected, does not lead to compression of the filter in time. In the framework of the present application, leaving a filter impulse response value that is not selected or not determined, neglected or not used corresponds to a simple modification to the coefficient of the delay operator z−1. (QMF filter bank) Does not lead to significant changes in the order of the individual addends of the polynomial. In other words, ignoring the filter tap or filter impulse response, leaving it, setting it to zero, or not paying extra attention to it, leads to a new distribution of filter taps with respect to the power of the delay operator z−1. No. The filter tap or filter impulse response value that is selected or determined following the filter impulse response value that is not selected or determined is not changed with respect to the power of the delay operator.

  In other words, the compressed subband filter impulse response as configured by the filter impulse response constructor 305 includes a zero value corresponding to the filtered tap of the unselected filter impulse response value. Or the compressed subband filter impulse response may not include any respective unselected filter impulse response values. In other words, the filter impulse response constructor 305, in principle, takes the same number of subband filter impulse response values as the input subband filter impulse response, except for values having an increased number of zero values. The compressed subband filter impulse response can be configured as, or the compressed subband filter impulse response is copied by the filter impulse response constructor 305 and ignores the unselected values As such, it can have a shorter overall length.

  Since the real-valued filter impulse response value leads to a significant complexity reduction compared to the complex-valued filter impulse response value, the filter-impulse response constructor 305 may provide some absolute value of the selected filter-impulse response value. The value can also be output advantageously. This mode of operation is particularly attractive in subbands corresponding to higher frequencies where human hearing is less sensitive to phase relationships.

  As a result, the subband impulse response value of the subband corresponding to the center frequency above the boundary frequency is the absolute value, imaginary part, real part, phase, linear combination, polynomial combination or at least one of the above-mentioned elements Can be arbitrarily replaced with a real-valued expression. Also, the imaginary part of a complex value is considered to be a real number in the framework described herein. Depending on the specific implementation, the boundary frequency may be in the range of 1 kHz to 10 kHz, but in many applications the implementation of the boundary frequency in the range between 1 kHz to 5 kHz or 1 kHz to 3 kHz is typical of humans. It can be used in consideration of auditory characteristics. Further, depending on the specific implementation of the filter compressor, the described replacement of the complex-valued filter impulse response value with a real-valued value based on the complex-valued filter impulse response value is selected or determined. It can also be implemented in response to filter impulse response values that have not been selected or have been determined. Alternatively or additionally, filter impulse response values belonging to the subband corresponding to the center frequency above the boundary frequency are usually replaced with corresponding real-valued values based on complex-valued filter impulse response values can do. In this context, using a determined or selected filter impulse response value may be based on such a filter impulse response value (e.g., an actual value) to replace the corresponding filter impulse response value. It should be noted that this includes the use of numerical values.

  FIG. 5 shows a further embodiment of a filter compressor 501 according to the present invention operating simultaneously on multiple filters. In FIG. 5, the different embodiments are outlined. Here, multiple filters (N filters denoted by ν = 0,..., (N−1)), where N is a positive integer, are input to the filter compressor 501 embodiment, Are input to the individual absolute value expression module 303, and the N expressions are input to the filter mask generator 502.

  Specifically, the embodiment of the filter compressor 501 shown in FIG. 5 is connected or coupled to a set of N filter converters 101 to which a set of real-valued time domain impulse responses Hν ( n, k) is supplied and, as explained previously, ν = 0,..., (N−1) is the index of the corresponding filter in the time domain. For example, in the case of a five channel input signal in the framework of a system like HRTF, for each of the five input channels, for each of the two headphone channels (left and right), the individual time domain filters are: Used to derive a total time domain filter of N = 10.

  In other words, the filter compressor 501 shown in FIG. 5 is provided with multiple sets of impulse responses, each set of multiple sets of filter impulse responses being different filter converters as illustrated in FIG. 101. However, with respect to a set of filter impulse responses as provided by an individual filter converter 101, a set of filter impulse responses includes L individual filter impulse responses, each of which is a specific number. Filter tap or filter impulse response value. As previously described in connection with the center frequency, each impulse response corresponding to an individual subband is associated with a center frequency, whereby the center frequency forms a plurality of center frequencies.

  Also, filter impulse responses that correspond to the same subband index k, but belong to different sets of filter impulse responses as indicated by index ν, correspond to the same center frequency. In other words, to each center frequency of the plurality of center frequencies (as defined by a set of filter impulse responses), at least prior to compression, (exactly) one in each set of filter impulse responses. Corresponds to the filter impulse response.

  Each filter converter 101 provides a set of complex-valued subband filter impulse responses Hν (n, k) for each time domain filter, which is the filter compressor shown in FIG. 501 embodiments. Each subband filter impulse response for N different time domain filters is provided to an individual absolute value representation module 303, which is an absolute value representation for each of the N time domain filters. Alternatively, an evaluation expression is provided to the filter mask generator 502. The absolute value representation module 303 can be selected from one of the other embodiments of the filter compressor of the present invention outlined in this application, as indicated by the same reference signs.

  This (joint) absolute value representation forms the basis for a single mask generation M (n, k) exactly as in the single filter mask generator 304 in the previous embodiment. In the case where a whitening step is performed, this can be performed for the individual absolute value representation module 303 or only once for the joint absolute value representation.

  In the context of FIG. 15, an embodiment of a filter compressor 501 is described, in which (spectral) whitening is performed individually for each filter ν = 0,..., (N−1). The The filter mask generator 502 in this embodiment creates a single filter mask M (n, k) for all filters based on the N absolute value representations of all filters. This is a major advantage of embodiments of the present invention because the filter mask generator 502 can take into account how the compression filters are combined in a later state. . Each original filter is input to a filter calculator 305, as outlined, after which the filter compressor provides N filters so that each filter calculator is provided with the same mask M (n, k). Create a new filter AWν (n, k).

  Compared to equation (11), the (joint) absolute value representation A (n, k) of equation (11 ′) is obtained by defining an equally distributed weighting factor ω (ν) = 1 / N. It can convert into the result of (11). In other words, the calculation of the absolute value representation according to equation (11) provides more flexibility so as to allow weighting of the perceptual importance of each filter indicated by the index ν. ) Represents a special form of absolute value representation.

  By using the same filter mask M (n, k) for each of the N individual filters in the time domain, an embodiment of the filter compressor 501 makes the N individual after the filter compressor 501 Even the subband post-processing leads to a resulting compressed subband filter impulse response having entries with associated impulse response values that do not have a corresponding selected impulse response value in one of the other filters. A set of compressed subband filter impulse responses can be created for each of the N filters. Compared to the mask generator 304 of the embodiment shown in FIGS. 3 and 4 with the mask generator 502 of the embodiment shown in FIG. 5, the mask generator 502 uses N individual filters only in the time domain. In spite of providing N input subband filter impulse responses for generating a single mask M (n, k) showing all of the N subband filter impulse responses. It is important to note.

  In a further embodiment of the filter compressor 501, different mask generators 502 can be used, which in principle use different schemes to provide a common evaluation representation for all N filters in the time domain. it can. In other words, apart from applying an average, as shown in equation (11), the individual evaluation representations can be weighted, for example, with respect to the associated subbands as provided by the absolute value representation module 303. Simply by summing each value, by linearly combining each value, or by using more complex combinations (eg, secondary or higher order combinations) of each value in the evaluation expression. Can be combined into one evaluation expression.

  As already explained before, binaural decoder 602 combines 10 (5 audio input channels for 2 audio output channels (stereo)) into 4 HRTF filters, which are connected to stereo input signal 603. Can be applied immediately. However, the HRTF filter relies on spatial parameters 604 that are provided to the binaural decoder 602 to render the binaural stereo signal 605. As mentioned above, HRTF filters, in particular, often serve as filter taps so that very complex interactions between binaural stereo output signals for the human ear and sound source must be modeled in many cases. Number of non-obvious, non-zero or non-zero subband filter impulse response values. Each HRTF filter can be significantly longer, for example, to efficiently model the environment being modeled and the room characteristics of other influences.

  In particular in this regard, embodiments of the filter compressor 501 can be efficiently applied to significantly reduce the computational complexity for the binaural decoder 602. By reducing the number of associated subband filter impulse response values considered in the binaural decoder 602 framework, the binaural decoder 602 can be implemented with less computing power, for example correspondingly The low energy consumption is ultimately derived so that the clock speed of the binaural decoder to be reduced can be reduced due to the low number of calculations in a given period. Alternatively, the binaural decoder 602 can be made smaller for the same reason, in principle so that the second processing core can be avoided.

  As outlined in more detail in connection with FIGS. 7-13, 192 (= 3 · 64) used to convert 10 time domain HRTF filters to complex QMF domain or complex subband domain HRTF filter in the time domain with 896 (= 14 · 64) filter taps, for example, filter converter 601 or rather 10 Each of the filter converters 101 forwards it to 64 individual subband filter impulse responses including 16 (= 14 + 3-1) filter taps. The resulting 1024 filter taps for each of the 10 time domain HRTF filters is such that the filter compressor 501 embodiment multiplies the total number of filter taps by, for example, 4 to 256 (= 1024/4). Unless used to reduce only a significant computational burden on the binaural decoder 602. Although this embodiment is based on a system that includes L = 64 subbands for each of 10 HRTF filters in the complex QMF or subband domain, in principle any number of L subbands Can be used.

  Possible solutions for filter converters and operation in the complex subband domain (QMF domain) before further embodiments of filter compressors and methods for making compressed subband filter impulse response filters are described The filter is described in more detail. However, before discussing the technical background in more detail, particularly for filter converters, the general concept of applying a digital filter (in the time domain or in the subband domain) to digital audio input should be described.

  FIG. 7 shows a possible solution for a filter or filter element 700, which is provided with a digital audio input. It should be noted that in principle the digital audio input can be both a time domain signal and a signal in the (complex) subband domain. The filter element provides a digital audio output at the output, which represents the filtered digital audio input in response to a filter definition signal or a respective filter impulse response signal.

  The filter converter 101 includes a complex analysis filter bank 710 as a central component where a corresponding filter impulse response signal is provided, as shown in FIG. The complex analysis filter bank 710 analyzes the impulse response signal of the filter in the time domain, which is forwarded to the QMF domain by filtering with a set of L analysis filters, after which any downsampling of the coefficient L is performed. Subsequently, L is also a positive integer, preferably greater than 1 and indicates the number of subbands in the complex analysis filter bank 710. The analysis filter is usually obtained by complex modulation of the prototype filter q (n), where n is a positive integer indicating the index in the array of data or the index of the value in the signal. The output of the filter bank 710 is composed of L subband signals and generally represents a filter characterized by its filter impulse response in the time domain of the complex QMF domain. Specifically, the output of the complex analysis filter bank 710 is a set of subband filter impulse responses that are provided to the filter element 700 to perform filtering of the speech input signal in the complex QMF domain. It can lead to perceptually indistinguishable differences in the audio output signal compared to direct filtering in the time domain.

  The basic design of the prototype filter q (n) and the complex-modulated analysis filter bank is outlined in more detail, followed by further details. Further, in the following, it is assumed that the number of subbands is fixed at L = 64. However, as explained previously, this is not a limitation of the embodiments of the present invention, but merely serves as a suitable example.

  FIG. 9 shows in more detail a possible solution of the complex analysis bank 710. Complex analysis bank 710 includes a plurality of L intermediate analysis filters 720 for each subband output by complex analysis bank 710. Specifically, each of the L intermediate analysis filters 720 is connected in parallel to a node 730 where a time domain impulse response signal is provided as an input signal to be processed. Each of the intermediate analysis filters 720 is configured to filter the input signal of the complex analysis bank 710 with respect to the center frequency of the respective subband. According to the center frequencies of the different subbands, each subband is denoted by a subband index or index k, and k is a non-negative integer, typically in the range of 0 to (L-1). . The intermediate analysis filter 720 of the complex analysis bank 710 can be derived from the prototype filter p (n) by complex modulation with the subband index k of the subband to which the intermediate analysis filter 720 is applied. Further details regarding the complex modulation of the prototype filter are described below.

  The sampling frequency of the signal output by the intermediate analysis filter 720, either directly by the intermediate analysis filter 720 or by any downsampler 740 (shown by the dotted line in FIG. 9) is reduced by a factor L. As described above, the downsampler 740 supplied to each subband signal output by the corresponding intermediate analysis filter 720 may perform downsampling in the framework of the intermediate analysis filter 720, depending on the specific implementation. It is optional so that you can also. In principle, downsampling of the signal output by the intermediate analysis filter 720 is not necessary. Nevertheless, the presence of an explicit or implicit downsampler 740 can increase the amount of data provided by the complex analysis bank 710 selectively by L times leading to significant data redundancy. This may be an advantageous option in other applications.

  FIG. 10 shows in more detail a possible solution of the subband filtering 750 and its interaction with the filter converter 101. Subband filtering 750 includes a plurality of intermediate filters 760, where one intermediate filter 760 is provided with a complex-valued respective subband signal provided to subband filtering 750. As such, subband filtering 750 includes L intermediate filters 760.

  The filter converter 101 is connected to each intermediate filter 760. As a result, filter converter 101 can provide a filter tap for each intermediate filter 760 of subband filtering 750. Further details regarding the filtering performed by the intermediate filter 760 are described in the further course of the application. Thus, the filter taps output by filter converter 101 and provided to different intermediate filters 760 form an intermediate filter definition signal.

  Furthermore, it should be noted that embodiments, solutions and implementations can include additional and / or optional delays for delaying any of the signals or subsets of signals, which are omitted in the drawings. Nevertheless, delays or delayers can be included in elements (eg, filters) shown or added as optional elements in all embodiments, solutions and implementations, depending on their specific implementation. .

  FIG. 11 illustrates a possible solution for the complex synthesis bank 770. Complex synthesis bank 770 includes L intermediate synthesis filters 780 from which L subband signals are provided. Depending on the specific implementation of the complex synthesis bank 770 prior to filtering in the framework of the intermediate synthesis filter 780, the subband signal is upsampled by L upsamplers 790, which increases the sampling frequency by a factor of L. Thereby reconstructing the sampled frequency of the subband signal. In other words, any upsampler 790 is provided with a subband provided to the upsampler 790 in such a way that the information contained in the respective subband signal is retained while the sampling frequency is increased L times. Reconfigure or reform the signal.

  Nevertheless, as already described in connection with FIG. 9, upsampler 790 is an optional component so that upsampling can be implemented in the framework of intermediate synthesis filter 780. Therefore, the step of upsampling the subband signal performed by upsampler 790 can be processed simultaneously in the framework of intermediate synthesis filter 780. However, if the downsampler 740 is not explicitly or implicitly implemented, the upsampler 790 need not be explicitly or implicitly implemented.

  Intermediate synthesis filter 780 is connected via an output to adder 800 that sums the filtered subband signals output by L intermediate synthesis filters 780. The adder 800 is further connected to a real part extractor 810 that extracts a real-valued signal or rather a (real-valued) time domain output signal based on the complex-valued signal provided by the adder 800 or Form. The real part extractor 810 calculates the absolute value of the complex value signal provided by the adder 810, for example, by extracting the real part of the complex value signal provided by the adder 810. Or by other methods of forming a real-valued output signal based on a complex-valued input signal.

  The second possible solution for the complex synthesis bank 770 shown in FIG. 12 differs from the first possible solution shown in FIG. 11 only with respect to the real part extractor 810 and the adder 800. Specifically, the output of the intermediate synthesis filter 780 is sent from each subband to a real part extractor 810 that extracts or forms a real value signal based on the complex value signal output by the intermediate synthesis filter 780. Connected separately. The real part extractor 810 is then connected to the adder 800, which is derived from the L filtered subband signals to form the real-valued output signal provided by the adder 800. Sum all real-valued signals.

  As previously described, FIG. 3 illustrates a possible choice of filter converter 101. The filter is assumed to be given by its impulse response. Looking at this impulse response as a discrete time signal, it is analyzed by L band complex analysis (filter) banks 710. The resulting subband signal output is then exactly the impulse response of the filter applied separately in each subband of the subband filtering 750 shown in FIG. In the case shown in FIG. 8, the filter definition signal provided to the filter converter 101 and its complex analysis bank or complex analysis filter bank 710 is an impulse response signal indicating the amplitude / frequency characteristics of the filter, which is a subband. • Transferred to the domain. Thus, the output of the complex analysis (filter) bank 710 for each of the L subbands represents the impulse response of the intermediate filter included in the subband filtering 750.

  The complex analysis bank 710 is derived in principle from the analysis bank for the audio output signal, but has different prototype filters and slightly different modulation configurations, the details of which are outlined in the following description. The length of the prototype filter q (ν) can be designed to be relatively small. Further, because of the downsampling by the coefficient L, the length of the subband filter is a coefficient of L smaller than the sum of the lengths of the predetermined time domain filter and the prototype filter q (ν).

In this application, a tap or value that is never zero is a tap or value that is ideally not equal to zero. Nevertheless, due to implementation limitations in the framework of this application, a value or tap that never becomes zero is a real or complex value with an absolute value greater than a predetermined threshold, eg, 10 ^−b or 2 ^−b . Is a tap or value, and b is a positive integer according to the specific implementation requirement. In digital systems, this threshold is preferably defined in binary systems (base 2), and the integer b has a predetermined value depending on the implementation details. Typically, the value b is 4, 5, 6, 7, 8, 10, 12, 14, 16 or 32.

  Complex modulated filter bank

  In equations (13) and (14), θ and ψ are used to filter the real-valued discrete-time signal x (n) into complex-valued subband signals and complex-valued subband signals dk (m). Represents a (constant) phase coefficient for reconstructing a real-valued output sample y (n) from. Prototype filter and constant phase coefficients θ and ψ to give a complete reconstruction y (n) = x (n) in the case of dk (m) = ck (m) where the subband signal is unchanged It is well known that you can choose. In practice, the full reconstruction characteristics apply up to the delay (and / or code conversion), but in the calculations that follow, this detail will be discussed in the PCT / SE02 / 00626 “Aliasing with Complex Exponential Modulated Filter Bank” section. As explained in the case of a pseudo-QMF type of design as in “reduction”, it is ignored by allowing the use of a non-causal prototype filter. Here, the prototype filter is symmetric p (−n) = p (n), and its discrete-time Fourier transform P (ω) is essentially zero outside the interval | ω | = p / L. Become. Complete reconstruction is also replaced with nearly complete reconstruction characteristics. For subsequent derivations, it is assumed that a complete reconstruction is maintained for simplicity of the description and that P (ω) = 0 for p / L <| ω | = p. Further, the phase coefficient is assumed to satisfy a state where ψ−θ is equal to an integer multiple of 4L.

  In highly sampled filter banks, changes to subband signals prior to synthesis usually lead to the introduction of aliasing artifacts. This is overcome here by the fact that oversampling by a factor of 2 is introduced by using a complex-valued signal. Even though the total sampling rate of the subband samples is the same as the sampling rate of the discrete time input signal, the input signal is real-valued, and the subband samples are complex-valued. As outlined below, the absence of aliasing allows for efficient time-invariant signal processing.

  Subband filtering in complex modulated filter banks.

Another case of interest is G _k (ω) = exp (−iω) which leads to G (ω) = exp (iLω), such as y (n) = x (n−L).

  Approximate a given filter response with subband filtering

The advantage of this procedure is that any given filter h (n) can be efficiently converted to an intermediate subband filter response. If q (n) has K _Q · L taps, then the time domain filter h (n) of K _H · L taps has subbands with (K _H + K _Q −1) taps. Converted to a domain filter (24), K _H and K _Q are positive integers. If K _Q is equal to 3 (L · K _Q = 192) and the impulse response of the time domain filter corresponds to a length of K _H · 64 (L = 64), then each intermediate subband filter 760 is It has an impulse response length of taps by K _H + 3-1 = K _H +2.

  Prototype filter design for filter converter

In the following, the determination of the multi-slot QMF representation (subband domain) of the HRTF filter is described. The filter transformation from the time domain to the complex QMF subband domain is performed in the filter converter 101 by the FIR filter. Specifically, the following description outlines a method for implementing a predetermined FIR filter h (n) of length _NH in the complex QMF subband domain.

The subband that filters itself is implemented by the intermediate filter 760 within the subband filtering 750. Specifically, subband filtering is another application of one complex-valued FIR intermediate filter g _k (l) for each QMF subband with index k = 0, 1,. Consists of. In other words, in the following description, a special reference is made in the case of L = 64 different subband signals. Nevertheless, this particular number of subband signals is not essential, and an appropriate equation is also given in a more general form.

A filter converter 101 that converts a predetermined time domain FIR filter h (n) to a complex subband domain filter g _k (l) includes a complex analysis bank 710. The prototype filter of the complex analysis filter bank 710 of the filter converter 101q (n) of length 192 (= 3 · 64) for the special case of L = 64 subband signals is in excess of equation (28). Produced by solving the determined system in the least squares method. The filter coefficient q (n) will be described in more detail later for the case of L = 64 subbands.

Note that in the framework of the present application under equations based equations, it is understood that additional delay (see l ₀ and n ₀ ) coefficients, the introduction of further coefficients, and the introduction of window functions or other simple functions. Should. In addition, simple constants, constant addends, etc. can be dropped. Also included are algebraic transformations, equivalent transformations and approximations (eg, Taylor approximation) that do not alter the result of the equation in any or significant manner. In other words, both slight modifications and transformations that essentially lead to the resulting match are included when the equation or expression is based on the equation or expression.

In this case, the integers k = 0, 1,..., 63 are also subband indices, and l = 0, 1,..., (K _H +1) is the resulting intermediate filter. An integer indicating 760 taps.

  There is a special addend of (−2) in equation (32) compared to equation (24) because equation (24) was developed without considering the loss of the filter. Actual implementation always introduces a delay. Thus, depending on the specific implementation, further delayering or delay can be implemented, which is omitted in the figure for simplicity.

In many cases, the system of linear equations (28) is over-determined. Nevertheless, it can be solved or approximated in a least squares manner with respect to prototype filter coefficients q (n). Solving the system of linear equations (28) in the least squares method leads to the following filter taps of prototype filter q (n) that satisfy the following relationship for integer n from 0 to 191.
q [0] = − 0.2029343380
q [1] = − 0.1980331588
q [2] = − 0.1929411519
q [3] = − 0.187767444222
q [4] = − 0.1822474011
q [5] =-0.1767730202
q [6] = − 0.17096286636
q [7] = − 0.1651273005
q [8] = − 0.1591756024
q [9] = − 0.1531140455
q [10] = − 0.1469560005
q [11] = − 0.1407020132
q [12] = − 0.1343598738
q [13] = − 0.1279346790
q [14] = − 0.1214308876
q [15] = − 0.1148523686
q [16] = − 0.10820244454
q [17] = − 0.1014939341
q [18] = − 0.09466991783
q [19] = − 0.08788500799
q [20] = − 0.0809381268
q [21] = − 0.07396444174
q [22] =-0.06692996831
q [23] = − 0.0598343081
q [24] = − 0.0526783466
q [25] = − 0.045454615388
q [26] = − 0.0381833249
q [27] = − 0.0308428572
q [28] =-0.0234390115
q [29] = − 0.0159703957
q [30] = − 0.0084535584
q [31] = − 0.0008319956
q [32] = 0.068418435
q [33] = 0.0145585527
q [34] = 0.0224107648
q [35] = 0.0303113495
q [36] = 0.0382934126
q [37] = 0.04636022959
q [38] = 0.0545155789
q [39] = 0.0627630810
q [40] = 0.07110686657
q [41] = 0.0795512453
q [42] = 0.0881007879
q [43] = 0.0967603259
q [44] = 0.01555369658
q [45] = 0.144301000
q [46] = 0.1234514222
q [47] = 0.3226049434
q [48] = 0.1418970123
q [49] = 0.1513433370
q [50] = 0.1609240126
q [51] = 0.706673535
q [52] = 0.1805909194
q [53] = 0.9060685753
q [54] = 0.2009635191
q [55] = 0.2113373458
q [56] = 0.2211163080
q [57] = 0.2330113868
q [58] = 0.2441343742
q [59] = 0.2554997664
q [60] = 0.2671158700
q [61] = 0.2790029236
q [62] = 0.2911152349
q [63] = 0.0.3650503350
q [64] = 0.90252575713
q [65] = 0.9103585196
q [66] = 0.91769777825
q [67] = 0.9245760683
q [68] = 0.9310214581
q [69] = 0.9370596739
q [70] = 0.9427143143
q [71] = 0.9480070606
q [72] = 0.9529578666
q [73] = 0.9575850672
q [74] = 0.61919056158
q [75] = 0.965953165
q [76] = 0.9696879297
q [77] = 0.97317373547
q [78] = 0.97664156119
q [79] = 0.9794139640
q [80] = 0.9821827692
q [81] = 0.90847315377
q [82] = 0.87070697790
q [83] = 0.8992030462
q [84] = 0.9991409728
q [85] = 0.9928893067
q [86] = 0.94494539395
q [87] = 0.99958401318
q [88] = 0.99970525352
q [89] = 0.9980952118
q [90] = 0.9989716504
q [91] = 0.99968847806
q [92] = 1.0002369837
q [93] = 1.0006301028
q [94] = 1.0008654482
q [95] = 1.0009438063
q [96] = 1.0008654482
q [97] = 1.0006301028
q [98] = 1.0002369837
q [99] = 0.99968847806
q [100] = 0.9989716504
q [101] = 0.9980952118
q [102] = 0.99970525352
q [103] = 0.99958401318
q [104] = 0.94494539395
q [105] = 0.9928893067
q [106] = 0.9911409728
q [107] = 0.8992030462
q [108] = 0.87070697790
q [109] = 0.9847473377
q [110] = 0.9821827692
q [111] = 0.9794139640
q [112] = 0.97664156119
q [113] = 0.97317373547
q [114] = 0.9696879297
q [115] = 0.965953165
q [116] = 0.91619056158
q [117] = 0.9575850672
q [118] = 0.9529575866
q [119] = 0.9480070606
q [120] = 0.9427143143
q [121] = 0.9370596739
q [122] = 0.9310214581
q [123] = 0.9245760683
q [124] = 0.91769777825
q [125] = 0.9103585196
q [126] = 0.90252757513
q [127] = 0.8941712974
q [128] = 0.2911752349
q [129] = 0.2790029236
q [130] = 0.2671158700
q [131] = 0.2554997664
q [132] = 0.2441343742
q [133] = 0.2330113868
q [134] = 0.2211163080
q [135] = 0.2113373458
q [136] = 0.2009635191
q [137] = 0.9060685753
q [138] = 0.1805909194
q [139] = 0.70667353517
q [140] = 0.1609240126
q [141] = 0.1513433370
q [142] = 0.1418970123
q [143] = 0.3226049434
q [144] = 0.1234514222
q [145] = 0.1444301000
q [146] = 0.05053496658
q [147] = 0.09676033259
q [148] = 0.0881007879
q [149] = 0.0795512453
q [150] = 0.07110686657
q [151] = 0.0627630810
q [152] = 0.0545155789
q [153] = 0.0466032959
q [154] = 0.0382934126
q [155] = 0.0303113495
q [156] = 0.0224107648
q [157] = 0.0145585527
q [158] = 0.068418435
q [159] = − 0.0008319956
q [160] = − 0.0084535584
q [161] = − 0.0159703957
q [162] = − 0.0234390115
q [163] = − 0.0308428572
q [164] = − 0.038183333249
q [165] = − 0.045454615388
q [166] = − 0.0526783466
q [167] = − 0.0598343081
q [168] =-0.06692996831
q [169] =-0.073964444174
q [170] = − 0.0809381268
q [171] = − 0.08788500799
q [172] = − 0.09466991783
q [173] = − 0.1014939341
q [174] = − 0.1008202454
q [175] = − 0.1148523686
q [176] = − 0.1214308876
q [177] = − 0.1279346790
q [178] = − 0.1343598738
q [179] = − 0.1407020132
q [180] = − 0.1469560005
q [181] = − 0.1531100455
q [182] = − 0.1591756024
q [183] = − 0.1651273005
q [184] = − 0.17096286636
q [185] = − 0.1767670202
q [186] =-0.1822474011
q [187] =-0.187767444222
q [188] = − 0.1929411519
q [189] = − 0.1980331588
q [190] = − 0.2029343380
q [191] = − 0.207676267137

  FIG. 13 shows a simplified block diagram of an embodiment of the filter compressor 102 that includes a processor 820 and a filter impulse response constructor 305 that are connected in series between the input and output of the embodiment of the filter compressor 102. Is done. An embodiment of the filter compressor 102 receives at the input a set of input subband filter impulse responses having filter impulse response values at the filter taps provided to the processor 820. The processor 820 examines the filter impulse response values from at least two of the input subband filter impulse responses, and further in conjunction with FIG. As described in the context of the whitening module 402, a filter impulse response value having a higher absolute value can be selected. Further, the processor 820 may not select at least one filter impulse response value that has a lower absolute value compared to the at least one selected filter impulse response.

  In other words, the processor 820 of the embodiment shown in FIG. 13 includes the functionality of the absolute value representation module 303 and the mask generator 304. The filter impulse response constructor or rather the filter calculator module 305 can construct a compressed subband filter impulse response using the selected filter impulse response value, and the compressed subband filter impulse response is selected It does not contain a filter impulse response value or zero value corresponding to the filter tap of the unfiltered filter impulse response value. As previously described, the filter impulse response filter 305 sets the unselected impulse response value to zero, or by copying only the selected impulse response value or not selected It should be noted that the compressed subband filter impulse response can be configured by some other means of ignoring the filter impulse response value.

  As a result, the embodiment of the filter compressor 102 can generate a compressed subband filter impulse response from the input in a subband filter impulse response having a filter impulse response value at the filter tap, as shown in FIG. Embodiments of the inventive method of making can be implemented. Making it in the context of a compressed subband filter impulse response can be similarly understood as generating or providing the compressed subband filter impulse response as a system or in a computer readable storage medium. it can.

  As shown in connection with the description of the whitening module 402 in FIG. 4, the evaluation expression A (n, k) or rather the absolute value expression Aν (n, k), which can be implemented according to equation (4), is white. The described method of digitizing or weighting is explained in more detail in connection with FIG. Accordingly, FIG. 14a shows a schematic diagram of an example filter characteristic 850 as a function of filter frequency in the time domain. Furthermore, FIG. 14a schematically shows an arrangement of the corresponding frequency bands 860-0,..., 860-4, which corresponds to a subband having an index k = 0,. Each of those frequency bands 860 corresponding to one of the subbands having a respective subband index k (using summarizing the reference signs as indicated earlier) is further characterized with respect to the center frequency. Which is shown as dotted lines 870-0,..., 870-4 in FIG. The center frequency and the frequency band of each subband are determined by the internal configuration of the complex modulated filter bank used in the filter converter 101. Specifically, prototype with center frequency by subband index k. The filter q (n) determines the corresponding frequency band of each subband, as seen, for example, in the case of equation (14). For example, if the corresponding complex-modulated filter bank p (n) or q (n) prototype filter is a low-pass filter for a subband with index k = 0, an exponential function in equation (14) Is converted to a bandpass filter for a higher subband index k = 1 for complex modulation as expressed by

  FIG. 14 b shows a schematic diagram of the input subband filter impulse response as provided by, for example, filter converter 101. Specifically, FIG. 14b schematically shows an evaluation representation A (k, n) for different subbands, shown as a set of arrows 880. For simplicity only, a set of three arrows 880 for each subband is shown in FIG. 14b for each subband 890-0,. As indicated by parentheses 900-0,..., 900-2, the five subbands 890-0,..., 890-4 are subbands 900-0, 900-1,. Arranged into three subgroups 900-2, the first subgroup 900-0 includes only the first subband 890-0 (k = 0), while the second and third subgroup 900 −1, 900-2 includes two adjacent subbands 890-1 and 890-2 and 890-3 and 890-4, respectively, with respect to the center frequency.

According to whitening or spectral whitening implemented in the framework of the whitening module 402 shown in FIG. 4, according to equation (4) for each of the subgroups of subband 900, the maximum value of the evaluation expression is determined, Further, as shown in FIG. 14c, thereafter, each of the evaluation expression values is subtracted to obtain a whitened evaluation expression A _W (k, n). As a result of subtracting the maximum value of the evaluation expression, for each of the subgroups 900, the maximum contribution of the evaluation expression is set to zero, as shown by dot 910 in FIG. 14c.

For the determination of the maximum value for each of the subgroups 900 according to equation (4), each of the subgroups of subband 900 includes at least one whitened evaluation representation value having a value of zero, The remainder of the evaluated evaluation expression value A _W (k, n) is less than or equal to zero. As a result, at least one value in each of the subgroups 900 is set to zero, thereby representing a maximum value, so that in each embodiment, each of the subband subgroups determined according to the psychoacoustic model is , In the process of compression of at least one filter impulse response value for each of the subgroups 900.

Thereby, in the process of spectral whitening as represented by equation (4), spectral weights or spectral energy have higher frequencies from subbands with lower center frequencies by applying a whitening scheme Transferred to subband. A direct comparison of FIGS. 14b and 14c also clearly shows this. While the evaluation representation value in subgroup 900-2 in FIG. 14b is significantly less than those in subgroup 900-1, the resulting whitened evaluation expression value in subgroup 900-2 after applying the whitening procedure Is significantly larger than at least some of the values of the evaluation expressions of subgroup 900-1. In this regard, it should be noted that subgroup 900-1 includes two zero-valued evaluation representation values as indicated by dot 910, which is evaluated as shown in subgroup Figure 14b. This is caused by the fact that the expression A (k, n) contains two identical maximum values. However, this is not a violation of the application of equation (4). Equation (4) ensures that at least one value of the evaluation expression for each subgroup is set to zero, thereby representing a maximum value in relation to the whitened evaluation expression A _W (kn). Just make it.

  FIG. 15 shows a further embodiment of the filter compressor 501, which can process more than one input subband filter impulse response Hv (n, k). The filter compressor configuration shown in FIG. 15 is very similar to that shown in FIG. 5 and is implemented only with respect to the fact that the absolute value representation module 303 includes an absolute value and logarithmic function module 401 and a whitening module 402, respectively. Unlike the form, it is shown and described further in connection with FIG. In addition, the filter calculation module or filter impulse response constructor 305 includes a filter decimator module 403 and a gain calculator 404, respectively, that can be implemented as optional components in connection with FIG.

The embodiment illustrated in FIG. 15 is further different from the embodiment illustrated in FIG. 5 with respect to the mask generator 502 for multiple filters. Specifically, the mask generator 502 of FIG. 15 includes an average calculation module 920, which, for example, produces individual (optionally whitened) absolute value representations Aν (n, k) according to equation (9). Based on this, a joint absolute value representation A (n, k) can be calculated. More specifically, in the framework of equation (9), an individual absolute value representation or evaluation for each of the filters Aν (n, k) for filters ν = 0,..., (N−1). The representation is replaced with the appropriate whitened evaluation representation A _W v (n, k) so that the whitened evaluation representation value is provided by the whitening module 402 to the average calculation module 920. In an embodiment of the filter compressor 501, for example, as shown in FIG. 15, the individual filter impulse response constructors 305 for the different filters ν = 0,. The number of filters provided at 501, which can be implemented as a single (overall) filter impulse response constructor 305 ', as shown by the dotted line in FIG. Specifically, depending on the specific implementation and technical circumstances, it may be desirable to implement a single filter impulse response constructor 305 ′ rather than N individual filter impulse response constructors 305. This may be the case, for example, at least in the framework of a filter impulse response constructor, where the ability to calculate is not an essential design goal or requirement. In other words, the embodiment 501 shown in FIG. 15 can be viewed as an embodiment in which the processor 820 and the filter impulse response constructor 305 ′ are connected in series between the input and output of each filter compressor 501.

  Furthermore, it should be noted that FIGS. 1-6, 13 and 15 can also be considered as flowcharts of respective methods with respect to the method embodiments and the methods implemented by the filter compressor 102, 501 embodiments. “Flow direction” should be included in the signal direction. In other words, the above figures reflect not only the different embodiments of the filter compressors 102, 501, but also the methods implemented by these embodiments and the generation of the compressed subband filter impulse response itself. Both method embodiments are shown.

  As such, embodiments of the present invention relate to a filter compressor in the subband domain, sometimes referred to as QMF (QMF = Quadrature Mirror Filter Bank), which is the head for multi-channel sound experience on headphones, for example. It can be used in the field of voice applications such as part transfer function (HRTF) filtering.

  The present invention relates to the problem of computational complexity using long filters in the QMF domain. The present invention selects coefficients that are most relevant in the time frequency representation of one or more filters, creates a filter mask that shows the most relevant filter coefficients, or coefficients that are not covered by the filter mask. By ignoring, it teaches a new way to reduce the required computation when applying filtering in the QMF domain.

  However, in the filter compressor 501 embodiment, the processor 820 is provided to the filter compressor 501 when examining and selecting a filter impulse response value for the compressed filter impulse response output by the filter compressor. It is not necessary to consider all filters. However, in this case, the filter compressor embodiment may be due to a compressed filter impulse response or one or more input filter impulse responses that were not taken into account in the framework of examining and selecting filter impulse response values. A plurality of compression filter impulse responses. This is desirable, for example, when one or more filters are not perceptually important to reduce further computational complexity, and these filters are considered when examining and selecting filter impulse response values. You are not required to be put in. This can be done, for example, when one or more filters do not have a significant amount of energy or volume. In these cases, the distortion introduced by not examining and selecting the filter impulse response values based on these filters can be accepted depending on the particular circumstances of these filters.

Some embodiments of the invention include the following features.
Transforming a time domain filter or some filters into a complex QMF filter representation, creating a time / frequency representation of the absolute value of the filter in the QMF domain, applying spectral whitening of the representation of the absolute value, 1 Creating a filter mask showing the desired filter coefficients giving a time / frequency representation of the absolute value of one or more filters; creating a new complex QMF filter containing the coefficients indicated by the filter mask

  Adjust the gain of the new filter or filters to obtain the same new filter gain as the original filter.

An embodiment of an apparatus for recalculating a complex QMF domain representation of a filter is:
Converting the time domain filter to a QMF domain representation;
Creating a filter representation of the QMF representation of the filter,
Creating a filter mask based on the representation of the QMF domain representation of the filter, and creating a new QMF filter based on the first QMF filter and the filter mask.

  Some embodiments of the present invention can solve the problem of high computational complexity of filtering long filters. They introduce a filter compressor that operates in the complex QMF domain. Thus, some embodiments of the present invention may provide a reduction in filtering computational complexity. Embodiments of the present invention can be implemented, for example, as a filter compressor, a method for generating a compressed subband filter impulse response, a computer readable storage medium, or a computer program.

  The embodiment of the filter compressor 102, 501 provides an opportunity to significantly improve the overall sound quality, even though the impulse response characteristics associated with many speeches have a fairly sparse time / frequency code. In many cases, longer contributions exist only at lower frequencies, and the effective duration is much shorter than the normal filter length for higher frequencies. Embodiments of the present invention can take advantage of these features, for example in the form of filter compressors.

  In addition, the compressed subband filter impulse response is a set of one or more subband filter impulses that approximate or represent a simultaneous time domain HRTF filter stored therein as provided by the filter compressor embodiment. It should be noted that the response can be stored on a computer readable storage medium. Compared to each set of HRTF-related filter impulse responses in the complex QMF domain, multiple compressed subband filter impulse responses stored in a computer readable storage medium typically have a shorter impulse response. However, it can be realized by having a lower number of respective impulse response values, or by a reduced number of non-trivial or non-zero filter taps or a combination of both.

For example, the corresponding HRTF filter function includes K _H filter taps in the time domain, and the compressed subband filter impulse response stored in the computer readable storage medium has subbands with L subbands. When intended to be used in a system, shorter impulse responses are typically included less than (K _H / L) for at least one subband filter impulse response. Preferably, the at least one subband filter impulse response includes less than (K _H / L-3) non-trivial or non-zero filter taps.

  Furthermore, if a computer readable storage medium stores more than multiple sets of subband filter impulse responses corresponding to a compressed time domain HRTF filter, the corresponding set of compressed subband filter impulse responses is , Including a common data pattern, the common data pattern indicating an impulse response value, which is an obvious value in at least some sets of subband filter impulse responses stored in a computer readable storage medium. The impulse response value having or lacking is shown. In other words, the common data pattern relates to a filter impulse response value that is not selected in more than one set of filter impulse responses, which is determined by the filter impulse response constructor as part of the filter compressor embodiment. Not used. Such a (similar) data pattern may be caused by the common filter mask M (n, k), for example as provided by the mask generator 502 for the multiple input filter Hv (n, k).

  In other words, a computer readable storage medium may include multiple filter impulse responses as well as a single set (compressed) subband filter impulse response for different subbands. Each of these sets of filter impulse responses, when viewed as a whole, contains a common data pattern, which is given by a corresponding impulse response value that is zero or missing at all. Each of these sets of filter impulse responses includes the same common data pattern stored on a computer readable storage medium. For example, in a set of filter impulse response values, if a particular value indicated by time or sample index n and subband index k is missing or has a zero value, another set of subband Impulse response values identified by the same sample of impulse response or time index n and the same subband index k are missing, have zero values, or have other defined values. In this connection, different sets of filter impulse responses are identified or indicated by their respective indices n, which can be any integer value, for example in the range 0 to (N−1). Often, N is also the number of filters.

  In other words, the above-described data pattern refers to the filter impulse response value, which is not selected in the meaning described in the context of the filter compressor 501 embodiment, for example. Thus, the data pattern is a different set of filter impulse response samples or time index n and subband index k, all identified by a respective filter index ν, all set to zero or all missing. And the index (n, k) with respect to and can be recognized or defined.

  The computer readable storage medium may include, for example, an HRTF related filter. Further, the multiple sets of subband filter impulse responses stored on the computer readable storage medium may be a set of filter impulse responses for the spatial audio system.

  It is important to note that a computer readable storage medium can in principle be any computer readable storage medium. Examples for such computer readable storage media are portable storage media such as, for example, floppy disks, CDs, CD-ROMs, DVDs or any other storage media, Information can be stored in a computer readable manner. Furthermore, built-in memory such as RAM (Random Access Memory), ROM (Read Only Memory), Hard Disk Memory, NVM (Nonvolatile Memory) or Flash Memory can be used. In other words, the computer-readable storage medium in the meaning of the present application is not only a portable storage medium but also a built-in storage medium. In addition, computer readable storage media refers to media that can change or change data or information as well as those memories that cannot change their data information.

  Thus, according to embodiments of the present invention, a computer readable storage medium may have a plurality of subband filter impulse responses stored in it and approximating a time domain head-related transfer function, The multiple subband filter impulse filter responses have a shorter impulse response compared to the time domain head-related transfer function.

  Depending on certain implementation requirements of the inventive method embodiments, the inventive method embodiments can be implemented in hardware or in software. This implementation is a digital storage medium, a computer readable storage medium having electronically readable control signals stored thereon that cooperate with the processor so that the method embodiments of the present invention are implemented. For example, using a disc, CD or DVD. As such, embodiments of the present invention are generally computer program products having program code stored on a machine-readable carrier, where the program code is the method of the present invention when the computer program is executed on a processor. It operates to carry out the embodiment. In other words, therefore, an embodiment of the method of the present invention is a computer program having program code for performing at least one embodiment of the method of the present invention when the computer program is executed on a computer. It is. In this regard, the processor may be formed by a programmable computer system, programmable computer, central processing unit (CPU), application specific integrated circuit (ASIC), processor or other integrated circuit (IC). Can do.

  Although the foregoing has been particularly shown and described with reference to specific embodiments of the present invention, it will be understood by those skilled in the art that various other changes in form and detail may be made without departing from the spirit and scope of the invention. Like. It should be understood that various changes can be made in adapting to different embodiments without departing from the generic concepts disclosed herein and further understood by the claims.

Claims (57)

A filter compressor (102, 501) for generating a compressed subband filter impulse response from an input subband filter impulse response corresponding to a subband, including a filter impulse response value at a filter tap, ,
In order to find a filter impulse response value having a higher value and at least one filter impulse response value having a value lower than the higher value, the filter impulse response from at least two input subband filter impulse responses A processor (820) for examining a value, and a filter impulse response constructor (305) for constructing the compressed subband filter impulse response using the filter impulse response value having the higher value , The compressed subband filter impulse response is
Does not include a filter impulse response value corresponding to a filter tap of the at least one filter impulse response value having the lower value, or a filter of the at least one filter impulse response value having the lower value. A filter compressor that contains zero-value filter impulse response values corresponding to taps.   The processor (820) for examining the filter impulse response value is configured to process a complex valued filter impulse response value, and the filter impulse response constructor (305) includes a complex valued impulse response. The filter compressor (102, 501) of claim 1 configured to process values.   The processor (820) is configured to look up the filter impulse response value based on an absolute value such that the higher value is a higher absolute value and the lower value is a lower absolute value. A filter compressor (102, 501) according to any of the preceding claims.   The processor (820) is configured to determine the filter impulse response based on an absolute value, a real part, an absolute value of the real part, an imaginary part, an absolute value or a phase of the imaginary part of the filter impulse response value. The filter compressor (102, 501) of claim 2, configured to examine a value.   The filter compressor (102, 501) of any preceding claim, wherein the processor (820) is configured to calculate an evaluation representation based on the filter impulse response value of the input filter response. .   The filter compressor (102, 501) of claim 5, wherein the processor (820) is configured to calculate the evaluation representation based on a psychoacoustic model or a model based on characteristics of the human ear.   The filter compressor (102, 10) according to claim 5 or 6, wherein the processor (820) is configured to calculate an absolute value based on the filter impulse response value to obtain the evaluation representation. 501).   The filter compressor according to any of claims 5 to 7, wherein the processor (820) is configured to calculate a logarithmic value based on the filter impulse response value to obtain the evaluation representation. (102, 501).

  The processor (820) weights the evaluation representation based on a psychoacoustic model based on a center frequency of the subband corresponding to the input subband filter impulse response to obtain a whitened evaluation representation. The filter compressor (102, 501) according to any one of claims 5 to 10, which is configured as follows.   The processor (820) is configured to weight the evaluation representation based on at least one subgroup of subbands, each subband belonging to at most one subgroup of subbands. A filter compressor (102, 501) according to any one of claims 5 to 11.   The processor (820) is configured to weight the evaluation representation based on at least one subgroup of subbands, each subgroup of subbands including at least one subband, and further each subband. 13. The filter compressor (102, 501) according to claim 12, wherein the band belongs to exactly one subgroup of subbands.   The processor (820) is configured to weight the evaluation representation based on at least one subgroup of subbands, each subgroup including at least two subbands having a respective center frequency of the subband. 14. The filter compressor (102, 501) according to claim 12 or 13, comprising only subbands having adjacent center frequencies with respect to a set of center frequencies of all subbands ordered according to.   The processor (820) is whitened with each subband or subgroup of subbands corresponding to one impulse response value and including a predetermined, compatible, programmable or constant value. 15. Filter filter according to any of claims 5 to 14, configured to weight the evaluation expression to obtain a whitened evaluation expression so as to include at least one value of the evaluation expression. Compressor (102, 501).   16. The filter compressor (102, 501) according to any of claims 5 to 15, wherein the processor (820) is configured to weight the evaluation representation individually for each subband.   The processor (820) determines a weighting for each subgroup of subbands or for each subband, and for each of the subbands of subbands or each of the subbands. 17. The evaluation representation of claim 5 to 16, wherein the evaluation representation is configured to weight the sub-band including subtracting the determined maximum value from a respective value of the evaluation representation for each of the subbands. Filter compressor (102,501) in any one.

  The processor (820) determines the weighting for each subgroup of subbands or for each subband and determines the maximum value for each subgroup of subbands. Dividing the respective value of the respective subgroup of subbands or of the evaluation representation of the respective subband by the maximum value or derived value or the determined maximum value for the respective subband. The filter compressor (102, 501) according to any of claims 5 to 16, wherein the filter compressor (102, 501) is configured to weight the evaluation expression to include:   The processor (820) may determine at least one filter impulse response value or subband of each subband having a higher value regardless of the absolute value of the impulse response value of the subband or subband subgroup. A filter compressor (102, 501) according to any of the preceding claims, configured to find at least one filter impulse response value of the group.   21. A filter compressor (102, 501) according to any of claims 5 to 20, wherein the processor (820) is configured to find the filter impulse response value based on the evaluation representation.   The processor (820) has a value lower than the higher value for at least one subband of less than 50% of the total number of respective filter impulse response values or for at least one subgroup of subbands. The filter compressor (102, 501) according to any of the preceding claims, wherein the filter compressor (102, 501) is configured to examine the filter impulse response value. The processor (820) is configured to find a filter impulse response value such that the number of impulse response values having a value lower than the higher value for at least one subband is less than K _Q , and The number of filter impulse response values of the filter in the time domain corresponding to the band filter impulse response is at least K _Q · L, L is the number of subbands, and K _Q and L are positive Filter compressor (102, 501) according to any of the preceding claims, which is an integer. The processor (820) is configured to examine the filter impulse response values such that the number of filter impulse response values for the at least one subband is less than or equal to (K _Q -3). The filter compressor (102, 501) according to 23.   Based on the evaluation expression or the whitened evaluation expression, the processor (820) is configured such that a predetermined, adjustable, constant or programmable number of impulse response values have the higher value. 25. A filter compressor (102, 501) according to any of claims 5 to 24, configured to examine the filter impulse response value.   The processor (820) is configured such that the predetermined, adjustable, constant or programmable number is greater than or equal to the number of subbands, or greater than or equal to the number of subgroups of subbands. The filter compressor (102, 501) according to claim 25.   The processor (820) may reduce the filter impulse response value, the corresponding value of the evaluation expression, or the corresponding value of the whitened evaluation expression below a threshold value if the corresponding value of the whitened evaluation expression is below a threshold value. The filter compressor (102, 501) according to any of the preceding claims, configured to find a response value.   The processor (820) includes at least one filter lower than the higher value when the value of the filter impulse response value is close to an aliasing level of a filter bank corresponding to the input subband filter impulse response. A filter compressor (102, 501) according to any preceding claim, configured to find an impulse response value.   The processor (820) is configured to provide a mask M (n, k) indicating the impulse response value having the higher value, where n is an integer indicating a sample or time index, and k is The filter compressor (102, 501) of any preceding claim, wherein the filter compressor is an integer indicating an index of the subband of the at least two input subband filter impulse responses.   The processor has the mask M (n, k) having a binary value indicating whether the impulse response value indicated by the integers n, k has a higher value or a lower value than the higher value. 30. The filter compressor (102, 501) of claim 29, wherein the filter compressor (102, 501) is configured to provide:   The processor (820) according to any preceding claim, wherein the processor (820) is configured to examine the filter impulse response value comprises selecting the filter impulse response having the higher value. Filter compressor (102, 501). The filter impulse response constructor (305) further provides a value lower than the higher value by providing a respective subband filter impulse response value or a value based thereon as the compressed subband filter impulse response value. Setting said impulse response value to zero,
Ignoring the impulse response value having a value lower than the higher value, and, if the filter impulse response value is a complex value, a complex value filter impulse response value having a value lower than the higher value By providing at least one real-valued value based on
The filter compressor (102, 501) of any preceding claim, configured to configure the compressed subband filter impulse response with the filter impulse response value having the higher value. ).   The filter impulse response constructor (305) uses the filter impulse response value having a higher value by at least one of obtaining and copying a respective impulse response value to use the compressed subband filter. A filter compressor (102, 501) according to any of the preceding claims, configured to constitute an impulse response.

  The filter impulse response constructor (305) includes a compression subband, such that at least one adjusted filter impulse response value includes a larger absolute value compared to the absolute value of the corresponding filter impulse response value. A filter compressor (102, 501) according to any preceding claim, configured to adjust at least one filter impulse response value of the filter impulse response.   The filter impulse response constructor (305) adjusts the impulse response value having a higher value by multiplying each impulse response value by a gain factor specific to the subband according to each subband. A filter compressor (102, 501) according to any of the preceding claims.   The filter impulse constructor (305) adjusts the impulse response value having a higher value by multiplying each impulse response value by a gain factor specific to the subgroup according to each subgroup of the subband. A filter compressor (102, 501) according to any of the preceding claims, configured to:

  The filter impulse constructor (305) may provide a complex-valued filter impulse as a corresponding compressed subband filter impulse response value when the complex-valued input filter response value corresponds to a center frequency above the boundary frequency. The filter compressor (102, 501) of any preceding claim, configured to configure the compressed subband filter impulse response value by providing a real value based on a response value. ).   The filter impulse constructor (305) is configured so that the real value based on the complex value filter impulse response value is a real part, an imaginary part, an absolute value, a phase, a linear combination based thereon, a polynomial based thereon, and a 41. The filter compressor (102, 501) of claim 40, configured to be at least one of a real-valued expression based thereon.   41. The filter impulse constructor (305) is configured to provide the real-valued value by replacing the complex-valued input filter response value with the real-valued value. The filter / compressor (102, 501) according to any one of 42.   43. The filter compressor (102, 501) according to any of claims 40 to 42, wherein the filter impulse constructor (305) is configured such that the boundary frequency is in the range of 1 kHz and 10 kHz. .   The filter impulse constructor (305) is based on the complex-valued input filter response value as a corresponding compressed impulse response value of the compressed filter impulse response when the filter impulse response value has a higher value. 44. A filter compressor (102, 501) according to any of claims 40 to 43, configured to provide said real value.   The filter compressor (501) is configured to generate a plurality of sets of compression filter impulse responses based on the plurality of sets of input filter impulse responses, and each of the compression filters of the set of compression filters and impulse responses. The impulse response corresponds exactly to one center frequency of the plurality of center frequencies, and one compression filter impulse of each set of the plurality of sets of compression filter impulse responses to each center frequency of the plurality of center frequencies. Accurately responding, each input filter impulse response of the set of input filter impulse responses accurately corresponds to one center frequency of the plurality of center frequencies, and each center of the plurality of center frequencies. Multiple sets of input filter impulses in frequency It corresponds exactly to the one input filter impulse response of each set of answers, the filter compressor according to any one of the preceding claims (502).   The processor (820) is configured to look up the filter impulse response value from at least two input subband filter impulse responses of at least one set of input filter impulse responses of the plurality of sets; 46. The filter compressor of claim 45, wherein 820) is further configured to find a filter impulse response value having the higher value from at least two sets of input filter impulse responses corresponding to the same center frequency. 501).   47. The filter compressor of claim 46, wherein the processor (820) is configured to find a filter impulse response value having the higher value from a full set of input filter impulse responses corresponding to the same center frequency. (501).   The processor (820) finds a filter impulse response value having the higher value of at least two sets of input filter impulse responses of a plurality of sets of input filter impulse responses corresponding to the same sample or time index n. 48. A filter compressor (501) according to claim 46 or claim 47, wherein n is an integer. It said processor (820) is configured to calculate the evaluation expression for the input filter impulse response of the respective set Aνn (n, k) or whitened evaluation expression A _W [nu (n, k), ν is an integer indicating the input filter impulse response of the set, n is an integer indicating a sample or time index, k is an integer indicating the index of the subband, and the processor (820) Calculates said evaluation expression A (n, k) based on at least two evaluation expressions Aν (n, k) or based on at least two whitened evaluation expressions A _W ν (n, k) 49. The filter compressor (501) according to any one of claims 45 to 48, further configured to:

  The filter impulse response constructor (305) is configured such that the filter impulse response value of the compression filter impulse response corresponding to the same center frequency and the same sample or time index is set to zero, and the plurality of sets of compression filters The compression filter of the impulse response, the plurality of sets of compressions so that if not included in the impulse response or if the corresponding filter impulse response value is a complex value, it is replaced with the respective real value value 51. A filter compressor (501) according to any of claims 45 to 50, configured to constitute the compressed subband filter impulse response of a filter impulse response. A method for creating a compressed subband filter impulse response from an input subband filter impulse response corresponding to a subband, including a filter impulse response value at a filter tap, comprising:
In order to find a filter impulse response value having a higher value and at least one filter impulse response value having a value lower than the higher value, the filter impulse response from at least two input subband filter impulse responses Examining a value; and configuring the compressed subband filter impulse response with the filter impulse response value having the higher value;
The compressed subband filter impulse response is
Does not include a filter impulse response value corresponding to a filter tap of the at least one filter impulse response value having the lower value, or a filter of the at least one filter impulse response value having the lower value. A method comprising a zero value filter impulse response value corresponding to a tap.   53. A computer program for performing the method of claim 52 when executed on a processor.   A computer readable storage medium having multiple sets of subband filter impulse responses stored therein, each set of subband filter impulse responses approximated to a time domain head related transfer function related filter And each filter impulse response of the time domain head related transfer function related filter is greater than the sum of the lengths of the subband filter impulse responses of each set of subband filter impulse responses, or Each filter impulse response of the time domain head related transfer function filter is the subband filter impulse response of each set of subband filter impulse responses when the filter impulse response values are complex values. Complex-valued Greater than the sum of the length of the filter impulse response values, a computer readable storage medium.   55. The computer readable storage medium of claim 54, wherein the plurality of sets of subband filter impulse responses includes a common data pattern.   The common data pattern includes at least two sets of subband filter impulse response subband filter impulse response zero or missing filter impulse response values, or the filter impulse response values are complex numbers. 56. The computer readable storage medium of claim 55, wherein if it is a value, the value is associated with a real value.   57. The computer readable storage medium of any of claims 54 to 56, wherein at least one set of subband filter impulse responses approximates a spatial audio filter.

Images